There is considerable debate over the level of size dimorphism and inferred social behavior of Australopithecus afarensis. Most previous studies have analyzed size variation in single variables or multiple variables drawn from single elements. These approaches suffer from small sample sizes, underscoring the need for new techniques that incorporate measurements from multiple unassociated elements, reducing the influence of random sampling on size variation in fossil samples. One such technique, the template method, has recently been proposed but is limited to samples with a template specimen and is sensitive to a number of assumptions. Here we present two new resampling methods that do not require a template specimen, allow measurements from multiple unassociated elements to be included in a single analysis, and allow for significance tests between comparative and fossil multivariate data sets with missing data. Using these new methods, multivariate postcranial size dimorphism is measured using eight measurements of the femur, tibia, humerus, and radius in samples of A. afarensis, modern humans, chimpanzees, gorillas, and orangutans. Postcranial dimorphism in A. afarensis is similar to that of gorillas and orangutans, and significantly greater than in modern humans and chimpanzees. Because studies in living primates have examined the association of behavior with dimorphism in body mass and craniodental measurements, not postcrania, relationships between postcranial dimorphism and social behavior must be established to make robust behavioral inferences for A. afarensis. However, the results of this and past studies strongly suggest behavioral and mating strategies differed between A. afarensis and modern humans. Am J Phys Anthropol, 2008. © 2007 Wiley-Liss, Inc.